The Architecture of a System for the Indexing of Images by Content

Abstract

This paper presents the architecture of an image database system which provides a platform for the accommodation of various algorithms for interactive and automatic indexing, storage, and retrieval of medical images by content. The system maintains a dynamic hierarchy of image classes. The class hierarchy is used to narrow down the search to images of the same modality, anatomical characteristics, etc. Each image is classified into an image class based on information supplied by the user or obtained from the image itself. An important feature of the system is its ability to support multiple image indexing by content methods, in the form of description types. During system installation, one or more description types are selected for each image class based on the inherent characteristics of the class. Then, for each description type, a content description of each image in the class is generated and inserted in the description database. In the case of a query, the system generates one or more descriptions of the query image, automatically or interactively, and searches the logical database for similar descriptions. The images whose descriptions match those of the query image are retrieved for browsing by the user. This system may also be used as a platform for evaluating image content description methods, since new methods can be added to it easily.